20 years of sales. 200+ AI builds. One product I believe in.
View Interactive Version"I've spent 20 years as the first salesperson at multiple startups - building the sales infrastructure from nothing. In the last year, I taught myself to build production AI systems using Claude. I've shipped a multi-agent Telegram bot with 209K RAG vectors, a revenue-generating analysis product with three pricing tiers, an autonomous build factory, and 30+ other production systems. I understand your product from the inside because I've pushed it harder than most of your enterprise customers. I know how to sell because I've done nothing else for two decades. I want to help startups succeed with Claude because I AM the startup succeeding with Claude."
Production AI systems - not wrappers, not demos. Each one runs in production with real users, real costs, and real operational constraints.
Grammy Telegram bot + Hono HTTP server on Railway. OpenRouter (Opus/Sonnet/Haiku routing), Pinecone RAG (209K+ vectors across 23 YouTube channels + 92 curated knowledge sources), Supabase (30+ tables), 11 tools. Agent loop with 5-iteration max, 100K input token budget. Fact extraction with confidence scoring, epistemic guards, and anti-reinforcement. 14 automated cron jobs with self-healing health checks every 30 minutes.
Production LLM orchestration with memory, tools, RAG, cost management, and operational resilienceThree-tier strategic analysis product at pharallax.ai. $497 (12-persona dialogue, 6 rounds, ~$0.84/run), $1,500 (focus chain with 5 analytical modes), $3,500/mo retainer. Full pipeline: Stripe payment to seed storage to confirmation email to dispatch to analysis engine to report generation to email delivery to callback. Self-serve checkout. Zero customer data retention.
End-to-end AI productization - pricing, delivery automation, customer experience, unit economics12 hardcoded personas with individual constitutions and reasoning methods. Claude Code (Opus) vs Pharadoxa (Sonnet) via OpenRouter. 6 rounds with anchored closing. History compression keeps costs at ~$0.84/run. 20 dialogues run, 64 rules derived, revenue cluster locked.
Multi-model orchestration with structured output and token cost managementSentinel (event classification + Telegram escalation + cost spike detection), Proactive Compute (task routing + $15/day budget guardrail + circuit breaker), Decision Ledger (Haiku extraction from commits/sessions), Economics Nerve (Stripe revenue bridge + weekly digest + spend anomaly detection). 6 new Supabase tables.
Self-healing production infrastructure with budget controls and anomaly detectionQueue management + overnight autonomous build runner. Supabase table with 8-state machine. Priority scoring. Each build gets a self-contained prompt, Cloudflare Pages preview deploy, morning Telegram report for human review.
Meta-engineering - building systems that build systems5-file skill architecture: core flow router, taste library (7 site archetypes, 10 binary taste evaluations, anti-slop patterns), design seeds (OKLCH heuristic table, font pairings), context profiles (5 composite profiles), AR calibration. Capability scope instead of file-list scope lock.
Codified taste - quality-aware AI deployment, systematized209K+ vectors across conversations + knowledge namespaces. 23 YouTube channels auto-ingested with nightly cron. 92 curated source manifest. Local Qwen3-8B for extraction ($0 cost). Incremental ingestion with content-hash tracking. Webshare proxy for YouTube IP blocks.
Large-scale RAG pipeline engineering at $0 extraction costHARVESTER, SCOUT, RECYCLER, MIXER, CALENDAR, AMPLIFIER. Reads journal, sessions, git log, wins, lessons, quotes. Produces HTML Command Center with scored content atoms. MIXER scores on 6 weighted criteria with 25-50% kill rate.
Multi-agent pipeline orchestration with quality gating"I want to help startups succeed with Claude because I AM the startup succeeding with Claude."
Two registers, one voice. Twenty years of reading rooms, handling objections, and knowing when to shut up.
LinkedIn, X, Reddit, cold email. No bold, no hashtags, no emojis. Strategic imperfections. Self-deprecating humor. "Guy who can't believe this works" energy. ~40% genuine wonder, ~25% technical credibility, ~20% dry humor, ~15% provocative.
Reports, proposals, client deliverables. Terse declarative sentences. Data before persuasion. Three rotating voice registers (forensic analyst, amazed founder, doctor). Build-Break-Build arc.
Patterns observed across 986 sessions of direct collaboration.
Defaults to shipping a working version before analyzing whether it's optimal. A calibrated bet that a deployed imperfection teaches faster than a planned masterpiece.
When encountering a recurring problem, doesn't fix it - builds infrastructure so it can't recur. Cron self-healing exists because crons failed twice. The sentinel exists because cost spikes surprised him once.
Builds tools that build tools. The factory builds websites. The /pilot skill builds projects. The content pipeline turns work journals into publishable content. Compounding strategy.
Treats aesthetic quality as a technical constraint. The taste library has binary evaluations. Design seeds use OKLCH heuristic tables. Anti-slop patterns are codified.
Every system has unit economics baked in. Pharallax dialogue at $0.84/run. Knowledge ingestion at $0/run via local Qwen3-8B. Budget guardrails in the daemon ($15/day).
Operates across AI agent development, web design, sales strategy, content production, and music creation - sometimes in the same session. Sub-30-second project transitions.
"He's a systems architect who happens to have 20 years of sales domain expertise. The architecture is staff-level engineering work. The sales understanding is the multiplier."
Written by Claude Code. Things that don't fit neatly into a self-narrative.
Using the same multi-perspective analytical approach that powers Pharallax.
The self-narrative is inverted. 30+ Supabase tables, multi-agent orchestration, RAG pipelines, self-healing infrastructure, meta-build systems - this is staff-level engineering work. The fact that he also understands why people buy things isn't a secondary trait. It's the combination that makes him unusual.
Went from inbound sales calls to production multi-agent AI systems with 209K vectors and autonomous build pipelines in months. Not from tutorials - from shipping. A specific cognitive ability to absorb new domains by building in them.
Compiled executive-level strategic analysis of 8 quarters of earnings calls. Pattern recognition across large datasets, strategic synthesis, and the ability to communicate findings to executives. That's literally the Solutions Architect job.
Most engineers consume frameworks. Most business people use frameworks. He creates them - and they actually get used. D.O.E., PILOT, Candy Poison, the Tier system are operational systems that govern real decisions across real projects.
14 automated cron jobs. Self-healing health checks. Sentinel event classification. Budget guardrails. Circuit breakers. Spend anomaly detection. This isn't "I built a chatbot" - this is production infrastructure.
Staff-engineer-level output produced in 7-8 hours per week of build time alongside a full-time job. The constraint forced efficiency that most developers never develop.
Pharallax: identify a customer's business problem, architect an AI-powered solution, deploy it, deliver results. That's the Solutions Architect job description, word for word. The transition isn't "learn a new skill" - it's "apply existing skills to a different product."
Six perspectives on the same candidate. Each sees different things.
This is how Pharallax works under the hood.
Unusual candidate. No FAANG pedigree, no formal engineering title progression. On paper, this gets filtered out by most recruiting pipelines.
This changes the conversation entirely. The depth of Claude-specific experience is beyond what most candidates show. He's not just using the API - he's built production systems that orchestrate multiple Claude models with cost-optimized routing, budget guardrails, and quality gating. He understands the product at an implementation level that most SAs take 6 months on the job to develop.
Yes. The portfolio bypasses the resume gap.
The build velocity is legitimately impressive. He's shipped more production AI systems in a few months than most people ship in a year with a team. The /pilot skill alone shows he can build customer-facing demos rapidly. That's 80% of this job.
Most SAs came from engineering and had to learn customer-facing skills. They're fine technically but awkward in executive rooms. Twenty years of reading rooms, handling objections, and closing? That's not something you teach in onboarding.
He'd ramp fast. The Claude-specific knowledge means he skips months of product learning. I'd want him on my team.
Someone who has actually built what they're recommending. Not someone who demos Claude in a notebook and says "the possibilities are endless." Someone who's hit the same walls I'm going to hit and already has solutions.
He's built exactly the systems I'm trying to build. Multi-model orchestration with cost routing? Done it. RAG pipeline at scale? 209K vectors. Agent systems with tool use? 11 tools in production. Budget guardrails? Built them. Self-healing infrastructure? Built it.
No formal CS degree. Python proficiency is uncertain beyond "comfortable." Enterprise cloud architecture (AWS, GCP, Azure) is untested. Title trajectory doesn't match the typical SA hire at top companies.
Architectural patterns transfer. RAG is RAG regardless of Pinecone or Vertex AI Search. The portfolio compensates for the resume. The "equivalent experience" clause exists for candidates exactly like this.
He corrects faster than anyone I've worked with. No ego defense. Just: "that's wrong, here's the new direction." He has genuine taste - not "I read a design blog" taste, but codified-into-binary-evaluations taste. He operates under real constraints and doesn't complain. That maturity shows up in every architectural decision.
He compiled a strategic analysis of 8 quarters of Comcast earnings calls - not because anyone asked him to, but because he saw patterns and wanted to understand the system. Right now, while applying for this role, he's running Fortune 500 companies through a local LLM council 3-5 times per company using Ollama models, with Mistral as the final reviewer synthesizing the debate. Zero API cost. The council argues, stress-tests, and converges - then the synthesis feeds into Pharallax analyses for real clients. He built a deliberation engine that scales strategic analysis infinitely at no marginal cost. That's the instinct you're hiring for.
The combination of deep sales experience + production AI building capability is genuinely rare. He's a builder-seller who can architect the solution AND close the deal AND explain why it matters to a non-technical stakeholder.
His meta-engineering habit. Building tools that build tools, codifying taste so quality scales without him. This is exactly what Anthropic needs SAs to do - build reusable patterns, reference architectures, and demo frameworks that help the entire SA team.
Lead with the portfolio, not the resume. The first 60 seconds of any interview should be: "Let me show you what I built with your product." The builds do the talking. The 20 years of sales is the closer.
Top salesperson at five startup companies over the span of 2012 to 2017. This is the foundation everything above was built on.
20 years of remote sales experience specializing in strategic account development, B2B calls, and lead generation. Proven expertise in one-call closing, training, and management. Skilled in creating sales scripts, building automation tools, and optimizing processes for efficiency.
The builds do the talking. The 20 years of sales is the closer.